The ZenML MCP server acts as a smart bridge that allows AI assistants to understand and interact with machine learning workflows. Think of it as a translator that lets an AI look into a project’s machine learning "factory" to see what is being built, which tools are being used, and whether everything is running smoothly. Instead of manually checking a dashboard, developers can simply ask their AI assistant for updates on their models and data pipelines. For those managing complex MLOps environments, this tool provides deep visibility into the ZenML ecosystem. It exposes a wide range of metadata, including pipeline configurations, stack components, and artifact details. The server can even retrieve specific step logs and source code, enabling an AI to help debug failed runs or explain the logic behind a specific part of the pipeline. It transforms the AI from a general-purpose chat bot into a specialized MLOps co-pilot that has real-time context regarding the entire development lifecycle. Technical integration is streamlined through the Model Context Protocol, allowing the server to plug directly into hosts like Claude Desktop or IDEs like Cursor. While it primarily offers read functionality for monitoring users, schedules, and service connectors, it also includes the capability to trigger new pipeline runs using existing templates. By combining containerized deployment options via Docker with the efficiency of the `uv` Python package manager, this MCP server provides a robust, secure way to bring sophisticated machine learning orchestration into a conversational development workflow.
Category: Data & Analytics
Tags: machine learning, mlops, orchestration, pipelines, zenml
uv installed locally (recommended via installer script or brew). * Clone the repository locally: bash git clone https://github.com/zenml-io/mcp-zenml.git Option A: Claude Desktop Bundle (Easiest) * Open Claude Desktop Settings. * Drag the mcp-zenml.mcpb file from the root of the cloned repository onto the menu. * Follow the prompts to add your ZenML server URL and API key. Option B: Docker * Pull the image: bash docker pull zenmldocker/mcp-zenml:latest * Run directly: bash docker run -i --rm \ -e ZENML_STORE_URL="https://your-zenml-server.example.com" \ -e ZENML_STORE_API_KEY="your-api-key" \ zenmldocker/mcp-zenml:latest ---uv) To configure the server for Claude Desktop or other MCP hosts using your local clone, use the following structure in your mcpServers config file. You must replace the dummy paths and credentials. json { "mcpServers": { "zenml": { "command": "/usr/local/bin/uv", "args": ["run", "/path/to/cloned/repo/server/zenml_server.py"], "env": { "LOGLEVEL": "WARNING", "NO_COLOR": "1", "ZENML_LOGGING_COLORS_DISABLED": "true", "ZENML_LOGGING_VERBOSITY": "WARN", "ZENML_ENABLE_RICH_TRACEBACK": "false", "PYTHONUNBUFFERED": "1", "PYTHONIOENCODING": "UTF-8", "ZENML_STORE_URL": "https://your-zenml-server-goes-here.com", "ZENML_STORE_API_KEY": "your-api-key-here" } } } }json { "mcpServers": { "zenml": { "command": "docker", "args": [ "run", "-i", "--rm", "-e", "ZENML_STORE_URL=https://your-zenml-server.example.com", "-e", "ZENML_STORE_API_KEY=your-api-key", "-e", "ZENML_ACTIVE_PROJECT_ID=...", "-e", "LOGLEVEL=WARNING", "-e", "NO_COLOR=1", "-e", "ZENML_LOGGING_COLORS_DISABLED=true", "-e", "ZENML_LOGGING_VERBOSITY=WARN", "-e", "ZENML_ENABLE_RICH_TRACEBACK=false", "-e", "PYTHONUNBUFFERED=1", "-e", "PYTHONIOENCODING=UTF-8", "zenmldocker/mcp-zenml:latest" ] } } }.cursor folder in the root of your repository. 2. Create an mcp.json file inside that folder using the configuration provided above. 3. Enable the ZenML server in Cursor settings. ---get_pipeline_runs and get_step_logs tools to identify a MemoryError in the 'train_model' step and suggests optimizing the batch size.Part of MCP Servers
The ZenML MCP server is an open source bridge that implements the Model Context Protocol to connect AI assistants to the ZenML API. It allows AI clients to inspect pipelines, runs, step logs, stacks, and registered models, as well as trigger new runs using snapshots.
The server works with standard MCP clients using stdio or Streamable HTTP transports, including Claude Desktop, Cursor, VS Code, Goose, and Claude Code. Experimental interactive MCP Apps specifically require Streamable HTTP support available in VS Code Insiders and Goose.
You can install it by configuring your client to run the local Python script via uv, pulling and running the official zenmldocker/mcp-zenml Docker image, or copying pre-configured connection snippets directly from the MCP Settings page inside your ZenML dashboard.
Yes, the ZenML MCP server is an open source project maintained on GitHub by ZenML. It can connect to both open source self-hosted ZenML deployments and managed ZenML Pro instances.
Yes, the server provides write functionality to trigger pipeline runs using the trigger_pipeline tool. It initiates runs using pre-configured frozen snapshots or legacy run templates.